There's nothing to fix in your project, and you're building again. We dug into the builder session like you asked. The failures hit five different commits, with and without NO_CACHE, and there's no variable or service called "worker" anywhere in your project. That name is a red herring: when a build's secret session drops mid-build, BuildKit reports whichever value it was fetching at that moment, and the "no active session" and "context canceled" errors on your retries came from the same drop. Railpack 0.37.0 isn't the culprit either, your passing builds run the same version. The drops were transient and stopped around 19:34 UTC on Aug 22. Your 19:43 deploy built cleanly on the same builder, passed its healthcheck, and is what's live now.